Why Algorithm-Generated Recommendations Fall Short.

The article discusses the limitations of algorithm-generated recommendations and the need for better ways to predict user preferences. Algorithms often rely on past behaviors to infer future preferences, leading to biases and inaccurate recommendations. The article provides examples of biases that affect decision-making, such as fast thinking, conflicting desires, and social norms. It suggests measures that organizations can take to improve algorithm design, including auditing algorithms for human bias, improving algorithm design to reflect users' normative preferences, training algorithms on different user data, and crafting algorithms that rely less on behavior and more on stated preferences. The article emphasizes the importance of investing in the behavioral science of algorithm design and moving beyond revealed preferences. [Extracted from the article]